Skip to content

level-3-citations.memory.md

Latest commit

 

History

History
171 lines (140 loc) · 5.27 KB

level-3-citations.memory.md

File metadata and controls

171 lines (140 loc) · 5.27 KB
id a1b2c3d4-5e6f-7a8b-9c0d-1e2f3a4b5c6d
type semantic
created 2026-01-20 10:00:00 UTC
modified 2026-01-22 15:30:00 UTC
namespace _semantic/decisions
title Adopt MIF for AI Memory Interchange
tags
architecture
memory-systems
integration
ai
aliases
MIF Adoption Decision
Memory Format Choice
temporal
valid_from valid_until recorded_at ttl decay access_count last_accessed
2026-01-20 00:00:00 UTC
2026-01-20 10:00:00 UTC
P730D
model strength
none
1.0
12
2026-01-22 15:30:00 UTC
provenance
source_type source_ref agent agent_version confidence trust_level
user_explicit
meeting:arch-review-2026-01-20
claude-3-opus
20240229
0.98
verified
embedding
model model_version dimensions source_text normalized
text-embedding-3-small
2024-01
1536
Decision to adopt MIF Memory Interchange Format for portable AI memory representation
true
citations
type title url role author date accessed relevance note
specification
Memory Interchange Format (MIF) Specification
source
@[[Robert Allen|Person]]
2026-01-23
2026-01-20
1.0
Primary specification document defining the format we are adopting
type title url role author date accessed relevance note
article
The Case for Portable AI Memory
supports
@[[Jane Smith|Person]], @[[John Doe|Person]]
2025-11-15
2026-01-18
0.92
Research article supporting vendor-neutral memory formats
type title url role author accessed relevance
documentation
Obsidian Help Documentation
background
@[[Obsidian Team|Organization]]
2026-01-19
0.85
type title url role author date relevance note
paper
Bi-temporal Data Models in Knowledge Systems
methodology
@[[Research Group|Organization]]
2024-08-20
0.78
Theoretical foundation for MIF's temporal model
type title url role author accessed relevance
repository
Subcog Memory System
extends
@[[Robert Allen|Person]]
2026-01-20
0.88
extensions
subcog
domain hash
architecture
sha256:7d8e9f0a1b2c3d4e5f6a7b8c9d0e1f2a

Adopt MIF for AI Memory Interchange

Context

Our AI systems currently use proprietary memory formats that are incompatible across different providers (Mem0, Zep, Letta, Subcog). This creates vendor lock-in and prevents memory portability when switching or combining AI memory systems. ^context

Decision

We will adopt the Memory Interchange Format (MIF) as our standard for AI memory representation and interchange. ^decision

Key Factors

  1. Dual Representation: MIF provides both human-readable Markdown and machine-processable JSON-LD formats
  2. Obsidian Compatibility: Direct integration with our existing knowledge management workflows
  3. Semantic Web Support: JSON-LD enables RDF tooling and semantic queries
  4. Local-First: No cloud dependencies, full data ownership
  5. Extensibility: Custom properties without breaking compatibility

Consequences

Positive

  • Memories portable between AI providers
  • Human-readable format for manual review and editing
  • Compatible with existing Obsidian vaults
  • Future-proof with semantic web standards

Negative

  • Migration effort for existing memory stores
  • Team training on new format
  • Tooling development for format conversion

Implementation Plan

  1. Phase 1: Implement MIF export from current Subcog format
  2. Phase 2: Build import adapters for Mem0, Zep, Letta
  3. Phase 3: Native MIF storage in new projects
  4. Phase 4: Deprecate proprietary formats

Relationships

  • supersedes [[proprietary-memory-format-2024]]
  • relates-to [[ai-infrastructure-standards]]
  • part-of [[platform-architecture-decisions]]
  • implements [[memory-portability-requirement]]

Entities

  • @[[MIF|Technology]]
  • @[[Subcog|Technology]]
  • @[[Obsidian|Technology]]
  • @[[JSON-LD|Technology]]
  • @[[Engineering Architecture Team|Organization]]

Citations

  • Memory Interchange Format (MIF) Specification by @[[Robert Allen|Person]] (2026)

    • Type: specification
    • Role: source
    • Relevance: 1.0
    • Primary specification document defining the format we are adopting. This is the authoritative source for MIF structure, conformance levels, and implementation requirements.

  • The Case for Portable AI Memory by @[[Jane Smith|Person]], @[[John Doe|Person]] (2025)

    • Type: article
    • Role: supports
    • Relevance: 0.92
    • Research article arguing for vendor-neutral memory formats in AI systems. Provides empirical evidence for productivity gains and reduced lock-in risks.

  • Bi-temporal Data Models in Knowledge Systems by @[[Research Group|Organization]] (2024)

    • Type: paper
    • Role: methodology
    • Relevance: 0.78
    • Theoretical foundation for MIF's temporal model, explaining valid time vs transaction time and decay functions for memory strength.